Engineering topics
Publications and source records attributed to Quirein, J. A..
The theory of classification methods and the functional steps in the manual training process used in the three phases of LACIE are discussed. The major problems that arose in using a procedure for manually training a classifier and a method of machine classification are discussed to reveal the motivation that led to a redesign for the third LACIE phase.
The Large Area Crop Inventory Experiment (LACIE) is a joint undertaking of the U.S. Department of Agriculture, the National Oceanic and Atmospheric Administration of the U.S. Department of Commerce, and the National Aeronautics and Space Administration. It is designed to verify an economically important application of remote sensing from earth orbital satellites. The first two phases of the experiment have been completed. A description of the experiment and a short discussion on the results and conclusions of the first two phases are presented. The LACIE design is compared with other designs for conducting crop inventories. An integrated design methodology (based upon accumulated LACIE experience) is discussed for conducting crop acreage surveys in developing countries.
A maximum likelihood estimation technique is used for the analysis of agricultural remote sensor data. The m-class probability of misclassification is estimated using unlabeled test samples and labeled training samples. A bound on the variance of a proposed unbiased estimator of the m-class probability of error is derived. The particular case in which each class density is assumed to be a mixture of multivariate normal densities is considered. The extension of spectral signatures in space and time is discussed.
The case is considered of n distinct, normally distributed classes or populations of two dimensional response vectors x = (lambda sub 1, lambda sub 2), where lambda sub i is a measurement of the relative reponse of x along channel i. The problem dealt with is to determine the best channel in the sense of divergence and in the sense of minimizing the probability of misclassification.
The problem is considered of determining a function F of the interclass divergence over all possible combinations of a fixed number of channels such that maximizing F will minimize the probability of misclassification.
The problem is considered of finding a dimension reducing transformation matrix B that maximizes the divergence in the reduced dimension for multi-class cases. A comparitively simple expression for the gradient of the average divergence with respect to B is developed. The developed expression for the gradient contains no eigenvectors or eigenvalues; also, all matrix inversions necessary to evaluate the gradient are available from computing the average divergence.
The problem dealt with concerns feature selection or reducing the dimension of the data to be processed from n to k. By reducing the dimension of the data from n to k, classification time is generally reduced. Yet the dimension reduction should not be so great that classification accuracy is impaired. Thus, the general problem is considered of classifying an n-dimensional observation vector x into one of m-distinct classes where each class is normally distributed with mean and covariance. It is shown that the probability of misclassification is minimized if a maximum likelihood classification procedure is used to classify the data. The dimension of each observation vector to be processed is conveniently reduced by performing the transformation y = Bx, where B is a K by n matrix of rank k. Thus, the n-dimensional classification problem transforms into a k-dimensional classification problem.
The B-average divergence for m-distinct classes, resulting from the linear transformation y = Bx, is proposed as a feature selection criterion, where B is a k by n matrix of rank k not greater than n. It is shown that if the B-average divergence resulting from B is large enough, then the probability of misclassification, considered as a function f the class of all k by n matrices, is essentially minimized by B. A computer program, utilizing a gradient procedure, is developed to numerically maximize the B-average divergence and results are presented for the Cl flight line. For this example, corresponding to 9-distinct classes, most of the discriminatory information is found to lie in a 3-dimensional subspace, defined by an appropriately chosen 3 by 12 matrix B.